In this issue:
- Auto-Paternalism—One limit on organizations is that "the buck stops here" is exhausting after a while—but effective delegation eventually means that the decisions that have to be made at the top of the org chart get harder. So effective founder-led organizations end up being built around their founders' mental limitations. This echoes the process of coming up with an investing strategy that's compatible with the personality of the person executing the strategy, but in the more quantifiable world of finance, the process has gotten more advanced.
- Growth—OpenAI has big losses and spectacular revenue growth—but that revenue growth actually understates how fast usage is increasing.
- AI Deployment—Watch the paywalls.
- S-Curves and TV—User-generated content is risky, but it's also a way to farm for talent.
- Increasing the Asymptote—The bigger the bundle, the more of a struggle it is to remind people of all the things they're paying for.
- China—A rally in equities spurred by trying to stop a rally in bonds.
Today's issue of The Diff is brought to you by Tegus—register today for a panel, moderated by The Diff's Byrne Hobart, on how AI tools are changing the way software developers work.
Auto-Paternalism
Markets are worth studying because they often illustrate broadly applicable truths in a way that, because markets are so narrow and quantifiable, wouldn't be visible in the messiness of reality. For example, it's intuitively true that there's some tradeoff between maximizing the average outcome and aiming for consistency—no one argues that Netflix is the peak of cinematic expression, that McDonald's is the endpoint of the culinary arts, or that Hilton has reached the theoretical limit of good hospitality, but they're all big and valuable businesses because their customers know what to expect and are likely to get it. In markets, we can go deeper, and talk about factor weightings—quantifying luck!—and we can see exactly how much investors value repeatability by looking at the fees hedge funds charge when they can plausibly offer it.
This even extends to fuzzy topics, like psychology. Coming up with an active investing strategy is actually a weird exercise in self-discovery. There are people who are motivated by a 50% drawdown, and people who lose their minds when they're losing money, and these people tend to run different strategies. There are investors with a naturally sunny disposition, who want to see the best in everyone, and they tend to do well in early-stage private investing and in growth stocks, though in both cases they'll have the occasionally embarrassing mistake. If you're surly, pessimistic, and convinced everyone is out to get you, that's bad news for your friends and romantic partners but great news for LPs in your short-focused fund.[1]
At the limit, managing a portfolio absorbs most of the portfolio manager's useful brain cycles, and that means that good portfolio management is a matter of crafting an information environment where the right news shows up at the right time, with minimal distractions. This is one reason stop-losses are such a common tool, even though it seems irrational to buy a stock at $10 and sell it because it dropped to $9. That is irrational, but the alternative is the much more irrational process of spending a disproportionate amount of your time re-researching and re-underwriting a position when the market is telling you that you do not understand what makes the stock go up and down.[2]
Companies go through the same kind of information optimization process. Managing any kind of organization means delegating decisions as much as possible, but still having the hardest ones flow upward. The classic line about this, from someone who ran a very complex organization indeed, comes from Barack Obama: "One of the first things I discovered as President of the United States was that no decision that landed on my desk had an easy, tidy answer. The black-and-white questions never made it to me—somebody else on my staff would have already answered them." His predecessor had a pithier summary: "I hear the voices and I read the front page and I hear the speculation... But I'm the decider, and I decide what's best."
This leads to one of those bizarre cognitive environments, where life is a series of unpleasant surprises of varying magnitudes. And since an organization is a mechanism for processing information and getting it to the appropriate party to make a decision, it's also, fundamentally, a system for simultaneously having an effective CEO and not subjecting this CEO to a nervous breakdown.
Some of this involves processes with the two-sided goal of 1) minimizing the odds of a sudden crisis, and 2) setting a baseline of performance that is as high as possible while being generally below what the founder would be motivated to provide.
Effectively delegating decisions does not mean making fewer hard decisions in a day. It actually means making more, because an organization where managers can effectively delegate is an organization that will grow faster, and come up with new and alarming problems. In your first week as a CEO, your big problem might be fixing some annoying AWS config issue. But with hard work and a healthy dose of luck, within ten years you might be testifying before congress about how your company isn't the biggest threat the country has faced since the Cold War!
And what that means is that founder-led companies have an org chart that is designed around the mental quirks of the founding team and their top executives. This is one reason the transition from founder-led companies to their first professional CEO is so challenging: that executive is stepping into a role that's designed around pushing somebody else to their limits, and will have a hard time keeping up. (The toughest transitions are often with notorious micromanagers; Jack Welch, for all his faults, was pretty obsessed with the details of GE's businesses, and Hank Greenberg at AIG apparently kept a lot of the details of the business in his head.) An organization that's perfectly optimized for one team will be utterly broken for anyone else, unless there turns out to be some Pareto-improvement successor CEO who happens to be equal to or better than the current one at every major responsibility (but who somehow didn't use that comparative advantage to start their own even more successful business).
Managing a company or a portfolio is partly a matter of managing emotional states. Which is one of those things that's easy to talk about until you try to make it concrete, because concretely it means having present-you take a paternalistic approach to managing future-you's emotional instability. But that's an essential part of the process: setting up an org chart or structuring a portfolio is, among other things, a special application of psychotherapy.
Short selling, of the catalyst-driven variety, is basically a convenient way to monetize conspiracy theories. The good news here is that some management teams really are out to get you, or are sufficiently incompetent that if they were trying to ruin their investors they wouldn't do anything differently. ↩︎
A corollary to this is that the nature of the stop loss is different depending on the kind of trade you're doing. If you're doing some kind of intraday strategy where you catch small moves, you'll tend to have tighter stop-losses, which loosen a bit if you have some thesis about how the market will reevaluate a given stock over the next few months. And longer-term investors seem to set a sort of fundamental stop-loss; if you own a stock because you think margins are expanding, then you might sell regardless of price if margins stop going up. Meanwhile, short sellers tend to structure their portfolios to deal with the fact that a company trading at 5x its real worth can easily trade up to 10x. They're either adjusting positions continuously or diversified enough that any one mistake won't kill them. ↩︎
A Word From Our Sponsors
AI tools like Microsoft Copilot are transforming software development, but how effective are they really? In this exclusive panel moderated we’ll dive into the reality of AI-enhanced coding. Experts from Microsoft, Adobe, and Cisco will share their experiences and offer actionable insights on the true impact of AI on developer productivity.
Elsewhere
Growth
OpenAI is projecting $3.7bn in revenue this year, with losses of around $5bn. That would be 18x the revenue of 2023, and they expect 213% growth next year. There are a few interesting things to note about this:
text-davinci-003
launched in late 2022 at a price of $0.02/1k tokens. GPT-4o is much more capable and was made available in May with a price per output token 25% lower (input tokens are a quarter of DaVinci's pricing). So that revenue growth actually understates how fast usage is increasing. One of the paradoxes of digital goods is that consumption increases are often invisible; almost everyone who logs hours a day in Instagram or TikTok was once someone with more intermittent usage.- Their margins are obviously not good, but if funding is abundant, growth happens fast, and scaling laws continue to hold, it's optimal for them to figure out how to effectively spend money as quickly as possible. If the funding market isn't asking a company to economize, it's not going to find the motivation to do so somewhere else.
- OpenAI's valuation has gotten much less demanding over time. 13X next year's revenue is just not that crazy for a company that plans to triple in size over the next year, even if the absolute amounts are intimidating.
Meanwhile, that round is slowly coming to a close, with Thrive leading, Microsoft and Nvidia potentially participating, and Apple out of the running ($, WSJ). Bigger, later-stage deals should have more strategic investors: they're better able to diligence the deal, they can make distribution or information rights part of the deal, and at this round size the list of potential investors who can actually swing the deal is short.
Disclosure: Long MSFT.
AI Deployment
Google is launching an AI-powered smarter smart reply feature, as part of several of its paid AI bundles. This is a good example of why it's so challenging for incumbents to win with new AI features (Edit: that was a typo—it's easier for incumbents to win, not harder): it's easier for Google to make Gmail 10% better than for someone to independently build something 110% as good as Gmail—and in AI, "as good as" also means having easy access to relevant information. (If your AI-based Gmail killer is good, but doesn't have access to the user's calendar, then its responses can't incorporate that information. Of course, it's just a few clicks to share calendar access with a new app, but another way to frame that is that Google has the structural advantage of saving a few clicks in onboarding, not to mention that Google is presumably a priority customer for Google, so if this connection ever breaks they'll be fixing it fast.) This also illustrates that AI services don't have the same gross margin structure as other kinds of software. There will probably be abundant, free AI tools, but the use cases where it makes sense to expensively deal with a large volume of tokens will be gated for a while, both to make the unit economics work and to allow a gradual rollout.
S-Curves and TV
When a media outlet has professionally produced content, it tends to be clustered around some quality level. There are some things that won't get signed off on, and others that are so good that the talent will prefer to go elsewhere. User-generated content has a broader continuum; plenty of garbage, but room for something excellent. As YouTube consumption has shifted towards traditional TVs, and as its monetization has also moved closer to what TV does, that presents a threat ($, Economist): YouTube is basically an ongoing R&D project for finding concepts and creators that are so compelling that they work with "iPhone and a tripod"-level production values, and can be scaled up from there. That generic user-generated content is generally not what advertisers want their brands to appear next to, though they'll tolerate it if the CPMs are low. What they really want is the inventory from when a small fraction of those creators realize they're on to something big, scale it up, and stick around on the platform that made it happen.
Increasing the Asymptote
More mature streaming businesses have a different set of constraints and opportunities. Amazon (disclosure: long) has mostly focused on on-demand streaming, but is moving into more live shows. Right now they're negotiating with Brian Williams to host an election night show. One of the problems with Amazon's bundle is that there are so many products that it's hard to get customers to try everything, which means it's hard to benefit from the cross-selling and churn reduction of a comprehensive bundle. (Every few weeks I'll get another plaintive email from them informing me that Amazon Music still exists.) In that setup, they have a comparative advantage in paying for splashy, event-based programs—the payoff is not that CPMs on election-night ads will be especially high, but that some fraction of customers will be downloading the Prime Video app, reviewing the other offerings, and giving Amazon one more way to show them ads and giving the customer one more reason to keep paying for Prime.
China
Chinese stocks have been rallying for the last week in response to Chinese local governments loosening restrictions on homebuying. There are two sides to this: the obvious one is that the deflating property bubble has been a big contributor to both slow economic growth and a longer-term bear case in China, in which their growth was too geared towards new construction than other kinds of consumption. On the other side of the balance sheet, though, what property selection offers the financial system is long-duration debt—one of China's immediate concerns was slowing a bull market in long-term government bonds, which was a hard-to-hide indication that local investors expected low growth in the future.
Diff Jobs
Companies in the Diff network are actively looking for talent. See a sampling of current open roles below:
- A company building ML-powered tools to eliminate software bugs is looking for a senior software engineer with a passion for Nix. (Washington D.C)
- An AI startup building tools to help automate compliance for companies in highly regulated industries is looking for a director of information security and compliance with 5+ years of infosec related experience. Experience with HIPAA, FedRAMP a plus. (NYC)
- A tech-bio startup creating the world’s largest database of protein-molecule interactions (and leveraging it to design novel drugs) is looking for a machine learning engineer with some experience training LLMs on distributed systems. (Remote)
- A team of former SpaceX, Air Force, and MIT engineers are looking for a senior mechanical design engineer to help mass-produce rapidly deployable nuclear microreactors to enable energy superabundance. (LA)
- A number of companies in The Diff network are looking for data scientists that love wrangling and analyzing alt. data to help inform investment decisions. (Remote)
- The policy development arm of an established stablecoin is looking for an analyst to help define US crypto policy from first principles. Some background in crypto or policy preferred, but not required. (NYC, Remote)
Even if you don't see an exact match for your skills and interests right now, we're happy to talk early so we can let you know if a good opportunity comes up.
If you’re at a company that's looking for talent, we should talk! Diff Jobs works with companies across fintech, hard tech, consumer software, enterprise software, and other areas—any company where finding unusually effective people is a top priority.